Quality comparison of 360° 8K images compressed by conventional and deep learning algorithms

Due to the accessibility of virtual reality in recent years, there has been a great interest in producing and streaming omnidirectional $(360 ^{\circ}$ field of view) high resolution images and videos. Since both high resolution and high quality are demanding for the storage and distribution of such content, the use of advanced compression methods is a key factor in achieving this goal. This paper provides an objective comparison of conventional image compression codecs (JPEG, JPEG XL, HEIC, AVIF, VVC Intra) and deep learning image compression algorithms with a JPEG AI framework recommendation. The visual quality evaluation is based on ten images from publicly available databases compressed to predetermined bit rates. Six full reference objective metrics (WS-PSNR, MS-SSIM, VIFp, FSIMc, GMSD, VMAF) are used to evaluate the visual quality of the compressed images. Modern image compression codecs outperform the oldest and most widely used codec JPEG in terms of bandwidth reduction but require more processing power and system resources.